1.011 AI not cycling GS Air units

[MechFO]

Now that GS is working against City Forts I decided to go back and see how many planes I could get to fly against Rostov.

I got around a 100 for 100% Mission setting and 150-200 for the 300% Mission setting, this out of 735 available and in range.

I noticed that even after attacking Rostov several times, the AI would keep allocating the same KG's over and over, even with nearly exhausted air base ammo. KG's at similar distance, with the same planes, loadout and plenty of ammo did not fly with 0% travelled.

I then started attacking other units around Rostov, again, the same few air units were committed.

As an example the Ju88 KG's at Gorlovka and Izyum are much beloved, the units at Stalino flew once, the ones at Voroshilovgrad and Krasnoarmeyskoe were not used at all, even though both places are closer to the battles than Izyum.

What the AI should maximally allocate to a single GS battle is a seperate discussion but at least it should try and cycle through available units, maybe based on % travelled. I don't think I've ever seen more than 150-200 bombers commited to a single GS event. Soviets don't have many Corps yet so 4 Rifle divisions is the biggest single concentration I've seen so far. I don't know if this means there is some hard cap at around 200 bombers or if more would be commited if there was more in there.

In the 65.1 save, switch 1st Panzer to A if you want to use them to test.

[Nix77]

There's something weird going on in the air war, especially with tactical bombers.

I've ran an AI vs AI game, T1-T60 v1.00.7, and T60-120 v1.00.11. Tactical bombers on both sides aren't doing really anything. Ok well they are, but seems like there's some kind of passivity. Part of the problem is the command structure that the AI can't handle. Southern Front doesn't have any air command in May'42 (could be due to Air Army upgrade?), and NW Front air command is at Black Sea, while half of the front is in north.

In May'42, Germans are close to Moscow, on Axis turn, only 3 IL-2 were shot down, and zero Stukas even though there were close to 30 ground battles. Meanwhile the VVS has 52 groups filled with Shturmoviks, and 2500+ sitting in the pool. G

Meanwhile the Germans didn't lose any Stukas on that attack round. They have 600 Stukas in pool, all squadrons almost full. No wonder, since they're placing all bets on 18th Army

This is on turn 49 so it should be all 1.00.7, no versions mixed yet on that round.

[Joel Billings]

I'll try to look into these issues. Nix do you have any saves to test out your issues?

As for the mission setting it is definitely not linear, and there is a reluctance to a very large number of planes in any one attack. There's a lot that goes into the routine, and the size of the enemy forces (and maybe attacking forces, can't remember now) has something to do with commitment levels. It's Gary's special secret formula.

[Nix77]

Oh, forgot to attach the save!

[Nix77]

Here's another save from the end of a Soviet AI vs AI victory, turn 186. The Soviets have over 20000 Shturmoviks in the pool, and only 64 air groups using them. I think the AI is using tactical bombers a bit too passively, or there's something preventing them from using them efficiently. The supply situation on the map is horrible though, most of the airfields seem to be void of ammunition and supplies


EDIT: The save contains few minor modifications, I tested lowering Soviets ARM production multiplier. Didn't seem to stop them from steamrolling!

[Joel Billings]

I used the save around Rostov to look into overuse of the same air groups. I do see the system is weighted to use the same groups, only using additional groups once the miles traveled started to go up for the original groups. Seems that once they got up to about 20% flown more groups came in, and then as these all went up in miles flown, more groups got used. I've added this to our list to look at, but I'd be surprised if there is a quick or easy change to be made here.

I do agree that in AI tests the losses to the IL-2s are lower than they probably should be. This could be an issue of the tac bombers generally having shorter ranges, and thus are probably less likely to be in a good position to attack. We could use some more results from human games so we know whether it's a issue with the AI or some other part of the game.